mlsanityData to 7 Oct 2026

Are Large Language Models Really Robust to Word-Level Perturbations?

arXiv 2309.11166 ↗First posted 20 Sep 2023cs.CL cs.AIcited by 32

On the map: LLM alignment / RLHF / preference optimisation → LLM preference optimization methods

Abstract (arXiv)

The swift advancement in the scales and capabilities of Large Language Models (LLMs) positions them as promising tools for a variety of downstream tasks. In addition to the pursuit of better performance and the avoidance of violent feedback on a certain prompt, to ensure the responsibility of the LLM, much attention is drawn to the robustness of LLMs. However, existing evaluation methods mostly rely on traditional question answering datasets with predefined supervised labels, which do not align with the superior generation capabilities of contemporary LLMs. To address this issue, we propose a novel rational evaluation approach that leverages pre-trained reward models as diagnostic tools to evaluate the longer conversation generated from more challenging open questions by LLMs, which we refer to as the Reward Model for Reasonable Robustness Evaluation (TREvaL). Longer conversations manifest the comprehensive grasp of language models in terms of their proficiency in understanding questions, a capability not entirely encompassed by individual words or letters, which may exhibit oversimplification and inherent biases. Our extensive empirical experiments demonstrate that TREvaL provides an innovative method for evaluating the robustness of an LLM. Furthermore, our results demonstrate that LLMs frequently exhibit vulnerability to word-level perturbations that are commonplace in daily language usage. Notably, we are surprised to discover that robustness tends to decrease as fine-tuning (SFT and RLHF) is conducted. The code of TREval is available in https://github.com/Harry-mic/TREvaL.

Key facts. Cited by 32 papers in the map (data through 7 Oct 2026). Placed in LLM preference optimization methods, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 37 of its 53 references that are in the map. First posted on arXiv 20 Sep 2023.

Builds on

37 of 53 references.

Closest ancestors · 1

Same direction, newest first

  1. 2022Training language models to follow instructions with human feedback4 Mar

Other references · 36

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2023LLaMA: Open and Efficient Foundation Language Modelscited by 13,958
  3. 2023Llama 2: Open Foundation and Fine-Tuned Chat Modelscited by 12,865
  4. 2023Judging LLM-as-a-Judge with MT-Bench and Chatbot Arenacited by 8,393
  5. 2017mixup: Beyond Empirical Risk Minimizationcited by 5,155
  6. 2018GLUE: A Multi-Task Benchmark and Analysis Platform for Natural Language Understandingcited by 4,854
  7. 2019HellaSwag: Can a Machine Really Finish Your Sentence?cited by 4,018
  8. 2021On the Opportunities and Risks of Foundation Modelscited by 3,961
  9. 2022Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedbackcited by 3,634
  10. 2017TriviaQA: A Large Scale Distantly Supervised Challenge Dataset for Reading Comprehensioncited by 2,895
  11. 2020Learning to summarize from human feedbackcited by 2,562
  12. 2015Character-level Convolutional Networks for Text Classificationcited by 2,293
24 more
  1. 2023A Survey on Evaluation of Large Language Modelscited by 1,227
  2. 2017Visualizing the Loss Landscape of Neural Netscited by 1,126
  3. 2019Unsupervised Data Augmentation for Consistency Trainingcited by 1,057
  4. 2023BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Datasetcited by 828
  5. 2019EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Taskscited by 800
  6. 2019Adversarial NLI: A New Benchmark for Natural Language Understandingcited by 798
  7. 2023Open Problems and Fundamental Limitations of Reinforcement Learning from Human Feedbackcited by 743
  8. 2023VoxPoser: Composable 3D Value Maps for Robotic Manipulation with Language Modelscited by 733
  9. 2023ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debatecited by 724
  10. 2019Is BERT Really Robust? A Strong Baseline for Natural Language Attack on Text Classification and Entailmentcited by 663
  11. 2023Bias and Fairness in Large Language Models: A Surveycited by 626
  12. 2020BERT-ATTACK: Adversarial Attack Against BERT Using BERTcited by 488
  13. 2021A Survey of Data Augmentation Approaches for NLPcited by 408
  14. 2023Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignmentcited by 394
  15. 2019Cosmos QA: Machine Reading Comprehension with Contextual Commonsense Reasoningcited by 281
  16. 2023Challenges and Applications of Large Language Modelscited by 251
  17. 2023On the Robustness of ChatGPT: An Adversarial and Out-of-distribution Perspectivecited by 208
  18. 2023LLM-Eval: Unified Multi-Dimensional Automatic Evaluation for Open-Domain Conversations with Large Language Modelscited by 100
  19. 2023When Large Language Models Meet Personalization: Perspectives of Challenges and Opportunitiescited by 89
  20. 2023TrustGPT: A Benchmark for Trustworthy and Responsible Large Language Modelscited by 56
  21. 2023Towards A Unified Agent with Foundation Modelscited by 46
  22. 2023LLM-Mini-CEX: Automatic Evaluation of Large Language Model for Diagnostic Conversationcited by 14
  23. 2023InstructEval: Systematic Evaluation of Instruction Selection Methodscited by 12
  24. 2023The instabilities of large learning rate training: a loss landscape viewcited by 3

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 2; 2024: 13; 2025: 11; 2026: 6.

32 papers.

Sort:

By area:

  1. 2024GSM-Plus: A Comprehensive Benchmark for Evaluating the Robustness of LLMs as Mathematical Problem Solverscited by 155
  2. 2024Fairness in Large Language Models: A Taxonomic Surveycited by 57
  3. 2024Is Your Model Really A Good Math Reasoner? Evaluating Mathematical Reasoning with Checklistcited by 48
  4. 2023Hijacking Large Language Models via Adversarial In-Context Learningcited by 41
  5. 2025Benchmarking Reasoning Robustness in Large Language Modelscited by 30
  6. 2024Not All LLM Reasoners Are Created Equalcited by 29
  7. 2024Step-On-Feet Tuning: Scaling Self-Alignment of LLMs via Bootstrappingcited by 14
  8. 2025Lifelong Safety Alignment for Language Modelscited by 13
  9. 2025UniGuardian: A Unified Defense for Detecting Prompt Injection, Backdoor Attacks and Adversarial Attacks in Large Language Modelscited by 12
  10. 2023Measurement in the Age of LLMs: An Application to Ideological Scalingcited by 11
  11. 2024Robustness of Large Language Models to Perturbations in Textcited by 11
  12. 2024QPO: Query-dependent Prompt Optimization via Multi-Loop Offline Reinforcement Learningcited by 10
20 more
  1. 2025Bias Beware: The Impact of Cognitive Biases on LLM-Driven Product Recommendationscited by 10
  2. 2024PPTC-R benchmark: Towards Evaluating the Robustness of Large Language Models for PowerPoint Task Completioncited by 9
  3. 2025Evaluating Robustness of Large Language Models Against Multilingual Typographical Errorscited by 9
  4. 2025RoboDriveVLM: A Novel Benchmark and Baseline towards Robust Vision-Language Models for Autonomous Drivingcited by 8
  5. 2024From Form(s) to Meaning: Probing the Semantic Depths of Language Models Using Multisense Consistencycited by 7
  6. 2025STEER-ME: Assessing the Microeconomic Reasoning of Large Language Modelscited by 7
  7. 2025Exploring Robustness of Multilingual LLMs on Real-World Noisy Datacited by 6
  8. 2024Online Safety Analysis for LLMs: a Benchmark, an Assessment, and a Path Forwardcited by 5
  9. 2025Position: Stop Acting Like Language Model Agents Are Normal Agentscited by 5
  10. 2024Interpreting token compositionality in LLMs: A robustness analysiscited by 4
  11. 2026Evaluating Robustness of Large Language Models in Enterprise Applications: Benchmarks for Perturbation Consistency Across Formats and Languagescited by 4
  12. 2025Evaluating Robustness of LLMs in Question Answering on Multilingual Noisy OCR Datacited by 2
  13. 2026Semantic Needles in Document Haystacks: Sensitivity Testing of LLM-as-a-Judge Similarity Scoringcited by 2
  14. 2024Securing Reliability: A Brief Overview on Enhancing In-Context Learning for Foundation Modelscited by 1
  15. 2026Where Do Prompt Perturbations Break Generation? A Segment-Level View of Robustness in LoRA-Tuned Language Modelscited by 1
  16. 2024Harmonic LLMs are Trustworthycited by 0
  17. 2025Pruning Cannot Hurt Robustness: Certified Trade-offs in Reinforcement Learningcited by 0
  18. 2026Guaranteed Jailbreaking Defense via Disrupt-and-Rectify Smoothingcited by 0
  19. 2026Validating LLMs in social science: Epistemic threats and emerging normscited by 0
  20. 2026Reliability Scaling Laws for Quantized Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2023Evaluating Concurrent Robustness of Language Models Across Diverse Challenge Sets
  2. 2025Mind the Confidence Gap: Overconfidence, Calibration, and Distractor Effects in Large Language Models
  3. 2024Look at the Text: Instruction-Tuned Language Models are More Robust Multiple Choice Selectors than You Think
  4. 2026Context-Length Robustness in Question Answering Models: A Comparative Empirical Study
  5. 2026Rethinking Faithfulness in LLMs: A Pairwise Context-Sensitive Perspective
  6. 2025Assessing Robustness to Spurious Correlations in Post-Training Language Models
  7. 2025A Study into Investigating Temporal Robustness of LLMs
  8. 2026LPDS: Evaluating LLM Robustness Through Logic-Preserving Difficulty Scaling